██████╗ ██╗   ██╗████████╗██╗  ██╗ ██████╗ ███╗   ██╗       █████╗ ███╗   ██╗ ██████╗██╗███████╗███╗   ██╗████████╗
██╔══██╗╚██╗ ██╔╝╚══██╔══╝██║  ██║██╔═══██╗████╗  ██║      ██╔══██╗████╗  ██║██╔════╝██║██╔════╝████╗  ██║╚══██╔══╝
██████╔╝ ╚████╔╝    ██║   ███████║██║   ██║██╔██╗ ██║█████╗███████║██╔██╗ ██║██║     ██║█████╗  ██╔██╗ ██║   ██║   
██╔═══╝   ╚██╔╝     ██║   ██╔══██║██║   ██║██║╚██╗██║╚════╝██╔══██║██║╚██╗██║██║     ██║██╔══╝  ██║╚██╗██║   ██║   
██║        ██║      ██║   ██║  ██║╚██████╔╝██║ ╚████║      ██║  ██║██║ ╚████║╚██████╗██║███████╗██║ ╚████║   ██║   
╚═╝        ╚═╝      ╚═╝   ╚═╝  ╚═╝ ╚═════╝ ╚═╝  ╚═══╝      ╚═╝  ╚═╝╚═╝  ╚═══╝ ╚═════╝╚═╝╚══════╝╚═╝  ╚═══╝   ╚═╝   
                                                                                                                   
██╗      █████╗ ███╗   ██╗ ██████╗ ██╗   ██╗ █████╗  ██████╗ ███████╗███████╗                                      
██║     ██╔══██╗████╗  ██║██╔════╝ ██║   ██║██╔══██╗██╔════╝ ██╔════╝██╔════╝                                      
██║     ███████║██╔██╗ ██║██║  ███╗██║   ██║███████║██║  ███╗█████╗  ███████╗                                      
██║     ██╔══██║██║╚██╗██║██║   ██║██║   ██║██╔══██║██║   ██║██╔══╝  ╚════██║                                      
███████╗██║  ██║██║ ╚████║╚██████╔╝╚██████╔╝██║  ██║╚██████╔╝███████╗███████║                                      
╚══════╝╚═╝  ╚═╝╚═╝  ╚═══╝ ╚═════╝  ╚═════╝ ╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚══════╝                                      

PythonAncientLanguages

📌 Project status update (Nov 30, 2025)
Development on this project will resume soon. Expect refactoring, dataset updates, and new model experiments.

Deep Learning for the recognition and interpretation of Ancient Egyptian Hieroglyphs, Sumerian Cuneiform, and other ancient scripts.

This repository explores the application of Deep Learning (Computer Vision + NLP) to detect, recognize and ultimately translate ancient written languages such as:

• Egyptian Hieroglyphs
• Sumerian Cuneiform
• Mayan Glyphs (future work)

The primary focus is Egyptian hieroglyphics, with the broader goal of building a general framework for ancient script recognition.

The project uses:

• Python 3
• TensorFlow / Keras
• Convolutional Neural Networks (CNNs)
• Traditional Computer Vision techniques where applicable

---

🔍 Overview

This project is inspired by the academic work of:

• Morris Franken (University of Amsterdam, 2013)
• His repository: https://github.com/morrisfranken/glyphreader

Franken’s work remains one of the strongest baselines for glyph recognition, but is limited by:

• small datasets
• lack of scalable annotation environments
• incomplete end-to-end pipelines

Other notable prior research includes:

• F. Gimbert’s project: https://github.com/fgimbert/Hieroglyphs (inactive)
• Barucci et al. (2021): High accuracy single-glyph classifiers
• Ubisoft / Google Arts “Fabricius” project
  • https://artsexperiments.withgoogle.com/fabricius
  • Dataset not publicly released

While classification accuracy for individual glyphs reaches 96% in recent work, real translation remains unsolved due to:

• glyph segmentation complexity
• directionality
• grammar rules
• missing corpora for supervised training
• few publicly available labeled datasets

This project aims to move beyond recognition into:

• segmentation
• classification
• interpretation
• transliteration
• (eventually) translation

---

🚀 Quickstart

Currently implemented scripts:

Feature extraction

python image_processing.py

CNN training

python train_model.py

Requirements

pip install -r requirements.txt

Python 3.x required.

---

🧭 Project Status

This repository is under active development again.

Current state:

• ✅ Feature extraction tool
• ✅ Basic CNN classification
• 🔄 Dataset curation ongoing
• 🔄 Model improvements planned
• 🔄 Refactor and modernization underway

Next milestones:

• Multi-glyph detection in real images
• Automatic segmentation
• Hieroglyph line parsing
• NLP pipeline integration
• Translation pipeline
• Annotator interface
• Cross-script generalization

---

🎯 Goals

• Glyph detection in real photographs
• Dataset tooling and augmentation
• Improved CNN / Transformer models
• Character-level recognition
• Sequence modeling
• Corpus integration
• Offline-first design (no cloud requirement)
• Public dataset building tools
• Open annotation environment

---

📦 Datasets

Public data is sparse and fragmented.

Known sources:

• Fayrose Middle Egyptian Dataset
  https://github.com/fayrose/MiddleEgyptianDictionaryWebsite

• Gimbert dataset
  https://github.com/fgimbert/Hieroglyphs/tree/master/hieroglyphs

• Google Arts Fabricius sample images
  https://github.com/googleartsculture/workbench/tree/main/src/assets/images

• JSesh database
  https://github.com/rosmord/jsesh

Note:
Google’s training dataset (~50,000 traced glyphs) is not released.

More info: See public_datasets.md.

---

🧪 Research References

• Franken et al. (2013)
• Domingo et al. (2017)
• Talwar (2017)
• Wiesenbach & Riezler (2019)
• Elnabawy et al. (2021)
• Barucci et al. (2021)

(Original list preserved)

---

🤝 Contributing

This is a research-driven repository.

Contributions welcome:

• datasets
• annotation tools
• segmentation ideas
• model architecture experiments
• OCR pipelines
• NLP approaches
• domain expertise

---

⚠️ Disclaimer

This project is experimental research.

No claims of correctness or historical interpretation accuracy are made.

This is not a commercial project.

---

📚 Resources

• Transfer Learning in CNNs
  https://machinelearningmastery.com/how-to-use-transfer-learning-when-developing-convolutional-neural-network-models/

---

“We decode the past using the tools of the future.”